{ "iq_scoring": { "title": "Tabularium AI IQ: Measuring What Official Records Must Get Right", "category": "Specification", "readTime": "4 min read", "date": "March 20, 2026", "categoryColor": "bg-indigo-100 text-indigo-800", "image": "", "description": "How Tabularium AI's Index Quality scoring model turns index completeness into a measurable, auditable number - and why that changes how AI output quality is enforced in official records workflows.", "content": "

Official records have no tolerance for invented facts. Every field in an index is either present in the document or it is not. Every value is either extractable from the source or it is absent. There is no legitimate middle ground where a model fills the gap from inference.

\n

IQ Scoring measures whether the document's required indexes have been fully extracted and structurally supported. Two independent evidence sources - one semantic, one structural - cross-check each other. The semantic source measures attribute presence, extraction confidence, and ambiguity resolution across each index field. The structural source measures whether extracted instances match the document's own anchors. When both agree, the score reflects that agreement. When they diverge, the divergence is diagnostic: it identifies exactly where extraction and document structure are in conflict - and flags it for targeted reprocessing rather than passing it forward.

\n", "documentViewer": { "containerId": "iq-pdf-viewer", "title": "Tabularium AI | Index Quality (IQ) Scoring Specification", "pageCountLabel": "8 pages", "pages": [ "assets/docs/iq_scoring_pages/page-01.png", "assets/docs/iq_scoring_pages/page-02.png", "assets/docs/iq_scoring_pages/page-03.png", "assets/docs/iq_scoring_pages/page-04.png", "assets/docs/iq_scoring_pages/page-05.png", "assets/docs/iq_scoring_pages/page-06.png", "assets/docs/iq_scoring_pages/page-07.png", "assets/docs/iq_scoring_pages/page-08.png" ] } }, "agent_core": { "title": "Tabularium AI Agent Core: Overview", "category": "Overview", "readTime": "3 min read", "date": "October 30, 2025", "categoryColor": "bg-teal-100 text-teal-800", "image": "", "description": "What the Agent Core is, how it connects to Tabularium AI, and the key components - Orchestrator, API Gateway, and adapters - that power automated indexing, fees, redaction, and more.", "content": "

The AI Agent Core is a free, open-source framework for rapidly building and deploying custom AI agents to automate Recording, eRecording, Indexing, Imaging, and Real Estate Analytics workflows, fully powered by the Tabularium AI platform. Deploy as a container, configure agent workflows via its REST/JSON API, and activate intelligent automation, such as indexing, fee calculation, image enhancement, and endorsement, without changes to your existing systems.

\n
\"Diagram
\n

Key Components

  • Orchestrator: Central workflow engine managing agent state and processing logic.
  • API Gateway: REST/JSON endpoints for triggering agent actions (index, calc, enhance, endorse, refine).
  • Webhook Dispatcher: Real-time status and result notifications.
  • Tabularium AI (TAI) Adapter: Secure access to Tabularium AI services.
  • Recording System (RS) Adapter: Bridge to recording and indexing environments.
  • Cloud Storage Adapter: Persists originals, enhanced images, and metadata.
\n

API Endpoints

  1. Indexing: Session + OCR/Indexing. Real Property: General/Party/Reference/Property & Legal + Semantic Enrichment. Vital: General + Vital.
  2. Refine: Reprocess with indexes, segment, or notes.
  3. Calc: Fee factors & distribution (Fiscal Enhancement=true).
  4. Provision: Index → Calc automatically.
  5. Redact: After indexing/refinement (Confidential Indexing=true).
  6. AutoRedact: Index → Redact automatically.
  7. Endorse: After indexing/refinement.
  8. AutoRecord: Index → Calc → Redact → RS record → Endorse.
" }, "agent_install": { "title": "Tabularium AI Agent Core: Azure Deployment Guide", "category": "Guide", "readTime": "4 min read", "date": "October 30, 2025", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "Step-by-step Azure deployment for Agent Core: resources, VNET, container Web App, configuration, endpoints, monitoring, and scaling.", "content": "

This guide details how to deploy Tabularium AI Agent Core on Microsoft Azure using container-based infrastructure. The Agent Core orchestrates document-processing agents and provides a secure REST/JSON API for official records automation workflows.

\n
1. Create Azure Resource Group
  1. Azure Portal → Resource Groups → + Add.
  2. Name, subscription, region.
  3. Create.
\"Azure Resource Group creation in Azure Portal (region/name filled).
\n
2. Provision Virtual Network (VNET)
  1. Create Virtual Network.
  2. Address space 10.0.0.0/16; subnets:
    • public: 10.0.1.0/24
    • private: 10.0.2.0/24
  3. Same region/resource group.
\"Azure Azure VNET configuration page (public/private subnets).
\n
3. Deploy Tabularium AI Agent Core (Container App)
  1. Resource Group → Create → Web App (Docker, Linux).
  2. Same region; Plan ≥ 2 vCPU / 3.5 GB.
  3. Image: tabularium/agent-core:latest.
  4. VNET integration: public subnet.
Screenshot: Web App (Docker) with image set.
\n
4. Configure Environment Variables
AGENTCORE_API_KEY=<your-api-key>
Screenshot: Configuration panel with variable set.
\n
5. Access and Use the API
GET  /v1/health\nPOST /v1/records/index\nPOST /v1/records/calculate
Screenshot: API response for /v1/health.
\n
6. Monitoring & Maintenance
  • Health checks: /v1/health
  • Logs: Monitoring → Log stream
  • Upgrades: Deployment Center
\n
7. Scaling & Custom Domains
  • Scale up/down plan.
  • Scale out: autoscaling.
  • Custom domains/certs under Settings.
" }, "recognition": { "title": "Recognition", "category": "Feature", "readTime": "4 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "AI-driven OCR and recognition that converts document files into page-aware, reviewable text, using selective page processing and document context to support indexing, verification, IQ analysis, and downstream understanding.", "content": "

Tabularium AI applies OCR and recognition as the first document understanding stage inside the indexing workflow. The objective is not just to read characters from page images, but to produce reliable, page-aware text in the right document context so downstream indexing, verification, IQ analysis, and metadata generation start from a stronger foundation.

1. Document Capture into Pages

Recognition begins by converting the source document into a validated page set. This ensures the workflow is operating on a real page structure rather than treating the file as an undifferentiated binary input.

2. Selective OCR Scope

Not every request requires the same OCR depth. Tabularium AI can process a focused subset of pages or expand to full-document recognition, depending on the requested recognition level. This allows the workflow to balance speed, coverage, and operational needs.

3. Context-Aware Page Recognition

Recognition is performed page by page, but with awareness of broader document context. Core pages can be interpreted together with surrounding context pages so the recognized text is more useful for downstream page analysis and document-level understanding.

4. OCR as Reviewable Text Output

Each selected page is processed through OCR to generate reviewable text output that preserves page order and document flow. This creates a usable text layer rather than isolated fragments, making downstream processing more consistent and reliable.

5. Scalable Recognition Pipeline

Recognition is designed for operational throughput across varied document sets. Pages can be processed in parallel, tracked consistently, and delivered as structured page-level text output for the rest of the workflow.

6. Workflow-Ready Foundation

The recognized text becomes the direct input for page processing, title determination, indexing, IQ analysis, and metadata generation. Recognition therefore serves as the OCR foundation that feeds the rest of document understanding.

" }, "indexing": { "title": "Indexing", "category": "Feature", "readTime": "4 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "AI-driven indexing that classifies document type, page purpose, and segment role, then converts recognized content into structured metadata for searchability, verification, and downstream automation.", "content": "

Tabularium AI applies indexing after recognition to transform page text into structured, document-aware metadata. The objective is not just to capture values, but to classify what the document is, determine how each page functions, and then generate indexed output in the right context.

1. Page Processing and Classification

Indexing begins with page-level analysis. Each page is evaluated to identify its purpose and segment signals, such as transaction, party clause, acknowledgment, exhibit, recital, endorsement, or confidential content. These page classifications determine how the page should be treated during indexing.

2. Title Processing and Document Identity

Indexing also determines the document header, including title, class, secondary titles, and explanation. This establishes the document identity that guides downstream interpretation and ensures indexing is shaped by the actual instrument, not just isolated page text.

3. Selective Page Indexing

Not every page is indexed the same way. Title pages, recordable pages, supporting pages, and blank pages each play different roles in the workflow. Tabularium AI focuses indexing effort where meaningful record data is most likely to appear while still preserving page-level context and page statistics.

4. Segment-Aware Index Generation

Once page purpose and document class are known, indexing runs in the appropriate segment context. This allows related content such as transaction data, party clauses, acknowledgments, confidential content, and other legal or business sections to be indexed with the right rules, aspects, and meaning.

5. Structured Metadata Assembly

Indexed results are normalized into consistent metadata with page association, field meaning, segment alignment, and document-level organization. This creates a cleaner output layer that is easier to review, verify, search, and use across operational workflows.

6. Workflow-Ready Output

The final indexing output becomes the structured metadata foundation for verification, IQ-based review, enrichment, downstream automation, and optional confidential handling. Indexing therefore combines classification and field capture into one document understanding stage.

" }, "verification": { "title": "Verification", "category": "Feature", "readTime": "4 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "AI-driven verification that reviews indexed results in gated, segment-level passes, applies bounded corrections, and works with Tabularium AI IQ to determine whether targeted reprocessing is required.", "content": "

Tabularium AI applies verification as a gated quality stage inside the indexing pipeline. The objective is not to re-create metadata from scratch, but to review indexed results, generate controlled corrections, and work with Tabularium AI IQ to decide whether deeper reprocessing is needed before downstream use.

1. Gated Post-Indexing Review

Verification begins only after indexing has produced structured metadata and only for the segments that were explicitly enabled for processing. This keeps review focused, controlled, and aligned to the business scope of the document rather than running as an open-ended global check.

2. Segment-Aware Verification

Verification is performed in focused legal and business segments such as transaction, party, acknowledgment, monetary, court, vital, or confidential content. Reviewing one segment at a time keeps the logic precise and the corrective output grounded in the right document context.

3. Completeness and Aspect Checking

Tabularium AI reviews whether expected indexed fields are present, whether captured values are appropriate, and whether each value has been assigned to the correct aspect. This helps detect missing indexes, incorrect classifications, and mismatches between value and meaning.

4. Patch-Based Corrections

Verification does not directly overwrite metadata. Instead, it produces bounded corrective actions such as adding missing data, updating incorrect values, or removing unsupported entries. This creates a controlled and reviewable refinement step rather than an opaque rewrite.

5. IQ as the Escalation Gate

After verification corrections are applied, Tabularium AI IQ evaluates the resulting metadata and segment quality. If the indexed result is strong enough, processing continues. If IQ detects meaningful issues, the document enters targeted refinement for the affected segments.

6. Two-Stage Quality Control

Together, verification and Tabularium AI IQ form a two-stage control system. Verification repairs obvious segment-level gaps early, while IQ determines whether the remaining result is strong enough to proceed or requires deeper reprocessing for higher-confidence output.

" }, "enrichment": { "title": "Enrichment", "category": "Feature", "readTime": "4 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "AI-driven enrichment that transforms indexed results into higher-level structured metadata such as parties, legal groups, chain records, and ownership history for stronger downstream analysis and use.", "content": "

Tabularium AI applies enrichment after indexing, verification, refinement, and correction to transform flat indexed results into higher-level structured metadata. The objective is not just to preserve individual indexed fields, but to organize them into richer domain objects that are cleaner, more interpretable, and more useful for downstream workflows.

1. Selective Enrichment Scope

Enrichment is not a single monolithic step. It runs only for the branches that are enabled and supported by the current document content. This keeps processing focused on the metadata that can be meaningfully elevated into richer structures.

2. Party Enrichment

Indexed party data is transformed into more structured party records with normalized values and clearer roles. This helps distinguish how parties function within the document and creates a cleaner party layer for downstream review, search, and analysis.

3. Legal Description Enrichment

Legal content is organized into structured legal groups rather than remaining as flat indexed values. This allows related legal elements to be grouped together, normalized, and summarized into a form that is easier to interpret and use operationally.

4. Chain and History Enrichment

Where the document supports it, enrichment derives higher-level chain and history records from the indexed and enriched context. This can include structured chain relationships, ownership progression, conveyance history, mortgage history, and other derived record views.

5. Structured Metadata Replacement

Enrichment does not simply append more raw data. It promotes indexed content into cleaner structured objects and replaces lower-level source representations where appropriate. This results in metadata that is more organized, more consistent, and easier to use across the platform.

6. Workflow-Ready Intelligence

The final result is a richer metadata layer that supports downstream analysis, review, history reconstruction, and operational workflows. Enrichment turns indexed fields into document intelligence that is more useful than flat values alone.

" }, "computation": { "title": "Computations", "category": "Feature", "readTime": "4 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "Rule-driven computation that derives fee factors, totals, and related fiscal outputs from the current metadata snapshot for accurate downstream financial and recording workflows.", "content": "

Tabularium AI applies computation after indexing, verification, and enrichment to derive fee factors, totals, and related fiscal outputs from the current metadata snapshot. The objective is not to interpret the document again, but to turn prepared metadata into accurate, workflow-ready financial results.

1. Metadata-Driven Calculation Scope

Computation begins from the current metadata layer rather than from document images or raw text. This allows the calculation stage to work from the latest indexed, enriched, and refined record view.

2. Unified Calculation Snapshot

The calculation input is broader than indexed fields alone. It can incorporate structured parties, legal information, counts, references, and other relevant metadata so fee logic is based on the full prepared record context.

3. Fee Factor Derivation

Tabularium AI derives fee factors from configured metadata signals such as page counts, party counts, references, penalty-related values, and other document-specific indicators. This creates a consistent basis for later fee and fund calculations.

4. Deterministic Financial Calculation

Once the fee factors are established, computation produces the fiscal result layer, including totals, itemized fees, and fund allocations. This step is designed to produce consistent financial outputs from the same metadata state.

5. Cleaned and Finalized Output

Computation filters and finalizes the result so the metadata reflects only meaningful financial values. The result is a cleaner output layer ready for recording, review, reporting, and downstream delivery.

6. Workflow-Ready Financial Metadata

The final output is an updated metadata snapshot containing derived totals, fee items, and related fiscal structure. Computation therefore serves as the financial derivation stage that turns prepared metadata into usable recording and operational results.

" }, "title_analysis": { "title": "Title Analysis", "category": "Feature", "readTime": "4 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "AI-driven title analysis that builds chain records, merges related document history across sessions, identifies defects and coverage gaps, and produces structured title-level reports with marketability-oriented scoring.", "content": "

Tabularium AI applies title analysis in two connected stages: first building chain and history records from individual document metadata, then aggregating those results into broader title-level chainsets for defect analysis and scoring. The objective is not just to list references, but to reconstruct title continuity, identify risks, and produce a more usable analytical title view.

1. Session-Level Chain Building

Title analysis begins by turning a document’s metadata into structured chain and history records. This includes the current record, referenced instruments, and other related title events needed to understand how the document fits into the larger chain.

2. Enriched Chain Context

Chain records are strengthened with related context such as legal information, document references, parcel details, plat references, and other supporting metadata. This creates a more complete chain view than flat index values alone.

3. History Reconstruction

Where supported by the metadata, title analysis also derives structured history such as conveyance, mortgage, and encumbrance activity. This helps place each record within the broader progression of ownership and title events.

4. Batch-Level Chain Aggregation

After individual session chains are built, Tabularium AI merges related records across a broader document set into title-level chainsets. This allows overlapping references, duplicates, and connected title events to be consolidated into a more coherent chain structure.

5. Defect and Coverage Analysis

The aggregated chainsets are then analyzed for issues such as chain breaks, missing links, authority gaps, unreleased liens, legal support gaps, and other title defects. This produces an examiner-style analytical layer rather than just a raw list of records.

6. Structured Title Reports

The final output combines chain records, history, analytical findings, completeness measures, breaks or gaps, and marketability-oriented scoring into a structured title report. Title analysis therefore turns metadata into a higher-level title view that is more useful for review, risk assessment, and downstream workflows.

" }, "visual_refinement_conversion": { "title": "Visual Refinement & Conversion", "category": "Feature", "readTime": "5 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "Automated visual refinement for official record images, including deskew, crop, contrast normalization, despeckle, image reversal, binarization, and TIFF-ready conversion for cleaner, more readable, system-ready output.", "content": "

Tabularium AI applies automated visual refinement to official record images before downstream OCR, indexing, QC, redaction, and delivery. The objective is to produce cleaner, more consistent, system-ready images with higher readability, lower noise, and better operational reliability across legacy and modern source formats.

1. Auto-Deskew

Scanned pages are automatically straightened when the source image is slightly rotated or fed off angle. This improves readability, supports more reliable OCR, and prepares the page for accurate boundary detection and further processing.

2. Auto-Crop

Automatic cropping removes unnecessary border areas, including common black edge artifacts from scanned and microfilm-derived images. This increases usable page area, improves visual clarity, and reduces file size for storage and retrieval efficiency.

3. Contrast Normalization

Documents with faded text, uneven exposure, or aging artifacts often require contrast correction. Automated contrast normalization balances light and dark regions across the page to improve legibility for both printed and handwritten content.

4. Despeckle and Noise Reduction

Background noise, speckling, and uneven page tone are automatically reduced to create cleaner document images. This removes visual interference while preserving the recorded content needed for review and machine processing.

5. Image Reversal for Photostats

Certain legacy photostat and inverted-source images require tonal reversal to restore normal document presentation. Automated image reversal converts dark-background, light-text images into standard light-background, dark-text output, improving readability and making the record more usable for OCR, indexing, and archival workflows.

6. Binarization and TIFF Conversion

Color and grayscale images can be processed into cleaner black-and-white TIFF output for official records workflows and land-record environments. This supports higher-quality conversion from difficult source material, including faded pages, bound volumes, and legacy imaging sources.

7. System Outcome

The result is a more uniform image set optimized for readability, archival consistency, downstream automation, and operational delivery into record-management and indexing workflows.

" }, "redaction": { "title": "Redaction", "category": "Feature", "readTime": "5 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "Automated redaction for official records using document classification, contextual field understanding, spatial reasoning, and jurisdiction-specific rules to apply precise, defensible redactions directly to the image.", "content": "

Tabularium AI applies automated redaction to official records using document understanding, field-level context, and jurisdiction-aware rules. The objective is not just to find patterns, but to make defensible redact-or-retain decisions and apply the correct redaction region directly to the image.

1. Document-Aware Redaction

Redaction begins with document classification. The same number, name, date, or identifier can require different handling depending on whether the record is a deed, lien, death certificate, court filing, permit, or another official instrument. Tabularium AI first determines what the document is, then activates the appropriate redaction policy for that record class.

2. Field Detection with Role Assignment

Pattern matching alone is not reliable enough for official records. Names, dates, addresses, IDs, and reference numbers often look similar at the text level but have different legal meaning. Tabularium AI evaluates surrounding language, labels, layout, and positional relationships to determine field type and role - for example, whether a value is a grantor, witness, notary detail, DOB, account reference, or other sensitive element.

3. Multi-Mode Redaction

Different fields require different redaction modes. In some cases only the value should be hidden. In others, both the label and value must be covered as a single protected region. For certain sensitive blocks, the correct action is to redact the broader section rather than an isolated token. Tabularium AI supports value-only, field-and-value, and region-level redaction based on document type, field type, and governing policy.

4. Spatial and OCR-Tolerant Interpretation

Official records often contain handwriting, degraded print, legacy imaging artifacts, and irregular formatting. Tabularium AI combines OCR output with spatial reasoning so that handwritten or weakly recognized values can still be associated with the correct printed label, field zone, or protected block. This improves reliability when canonical formats are missing or OCR confidence is uneven.

5. Jurisdiction-Driven Rules

Redaction behavior is driven by configurable policy rules tied to document class, field class, and jurisdiction. This allows the system to apply different handling for categories such as personal identifiers, notary details, witness information, medical or death-related content, protected victim data, and other sensitive elements where local or domain-specific requirements differ.

6. Reviewable, Defensible Output

The final output preserves the usability of the public record while protecting the data that must not remain visible. Each redaction decision is tied to document context, field interpretation, and policy logic, producing image output that is more consistent, more reviewable, and more defensible than pattern-only masking.

" }, "exchange_storage": { "title": "Exchange Storage", "category": "Subscription", "readTime": "4 min read", "date": "March 18, 2026", "categoryColor": "bg-blue-100 text-blue-800", "image": "", "description": "Exchange Storage for secure document intake, artifact persistence, unique output generation, and controlled result access across indexing, reporting, and downstream workflows.", "content": "

Tabularium AI uses a Exchange Storage layer to support document intake, workflow artifacts, generated outputs, and downstream delivery. The objective is not to add business logic, but to provide a consistent storage foundation for reading source content, writing results, and exposing workflow outputs in a controlled way.

1. Unified Storage Foundation

Exchange Storage provides a common layer for handling workflow inputs and outputs across the platform. This creates a consistent way to manage source files, generated metadata, reports, and other processing artifacts.

2. Input Content Access

Workflows can load source documents and previously stored artifacts from Exchange Storage as needed. This allows processing stages to start from stored content rather than requiring direct file transfer into every step of the pipeline.

3. Output Persistence

Generated results can be written back into Exchange Storage for later retrieval, delivery, or downstream use. This includes workflow artifacts such as metadata snapshots, reports, and other generated outputs.

4. Unique Output Generation

When workflows produce result files that must remain distinct, Exchange Storage supports unique output naming so artifacts can be stored safely without collisions. This is especially useful for asynchronous processing, callbacks, and repeated runs.

5. Controlled Result Access

Exchange Storage supports controlled access to generated results so outputs can be retrieved by downstream systems in a structured and secure way. This enables delivery flows that depend on stable result locations and workflow-ready output references.

6. Workflow Integration

The storage layer supports both inbound and outbound workflow movement. Documents can be loaded into processing stages, and finished outputs can be persisted for callbacks, client retrieval, reporting, and operational delivery.

7. Operational Reliability

By centralizing storage access behind one shared layer, Tabularium AI keeps document and artifact handling more consistent across workflows. Exchange Storage therefore acts as the common persistence layer for input content, generated outputs, and downstream result delivery.

" } }